This very likely uses network speech recognition, so most things (except for the wake-up word) are being streamed up and recognized in Amazon's data center (Siri, SVoice, etc all do this also)